Programmed cell death-index (PCDi) as a prognostic biomarker and predictor of drug sensitivity in cervical cancer: a machine learning-based analysis of mRNA signatures

Purpose: Cervical cancer is a significant public health concern, particularly in developing countries. Despite available treatment strategies, the prognosis for patients with locally advanced cervical cancer and beyond remains poor. Therefore, an accurate prediction model that can reliably forecast prognosis is essential in clinical setting. Programmed cell death (PCD) mechanisms are diverse and play a critical role in tumor growth, survival, and metastasis, making PCD a potential reliable prognostic marker for cervical cancer. Methods: In this study, we created a novel prognostic indicator, programmed cell death-index (PCDi), based on a 10-fold cross-validation framework for comprehensive analysis of PCD-associated genes. Results: Our PCDi-based prognostic model outperformed previously published signature models, stratifying cervical cancer patients into two distinct groups with significant differences in overall survival prognosis, tumor immune features, and drug sensitivity. Higher PCDi scores were associated with poorer prognosis. The nomogram survival model integrated PCDi and clinical characteristics, demonstrating higher prognostic prediction performance. Furthermore, our study investigated the immune features of cervical cancer patients and found that those with high PCDi scores had lower infiltrating immune cells, lower potential of T cell dysfunction, and higher potential of T cell exclusion. Patients with high PCDi scores were resistant to classic chemotherapy regimens, including cisplatin, docetaxel, and paclitaxel, but showed sensitivity to the inhibitor SB505124 and Trametinib. Conclusion: Our findings suggest that PCD-related gene signature could serve as a useful biomarker to reliably predict prognosis and guide treatment decisions in cervical cancer.

2 Supplementary Tables and Figures   2.1 Supplementary Tables Table S1.Genes associated to fifteen types of programmed cell death (PCD) patterns Table S2.Identification of differentially expressed genes (DEGs) by three different algorithms Table S3.Development of the PCD-related prognostic signature by machine learning algorithms Table S4.Comparison of the PCD-related prognostic signature and previously published signatures Figure S2.Sample inclusion and exclusion criteria of this study.Figure S3.Univariate Cox regression analysis identified 27 PCD-related candidate genes.Figure S4.Comparison of expression levels of 12 PCD-related genes in tumor and normal tissues.Figure S5.KM estimates of OS in cervical cancer patients with different expression levels of 10 PCD-related genes.Figure S6.Association between programmed cell death-index (PCDi) score and clinical features in cervical cancer patients.Figure S7.KM estimates of OS in patients with different PCDi scores for a particular FIGO cancer stage.Figure S8.KM estimates of OS in patients with different PCDi scores and FIGO cancer stage.Figure S9.Multivariate Cox regression analysis in different study cohorts.Figure S10.The expression profiles of different tumor immune infiltrates between PCDi-High and PCDi-Low groups through ssGSEA algorithm.Figure S11.Violin and scatter plots displaying the association between PCDi and T cell dysfunction potential and exclusion potential.

Figure S1 .
Figure S1.Kaplan-Meier (KM) estimates of overall survival (OS) in cervical cancer patients with different FIGO cancer stages.FigureS2.Sample inclusion and exclusion criteria of this study.FigureS3.Univariate Cox regression analysis identified 27 PCD-related candidate genes.FigureS4.Comparison of expression levels of 12 PCD-related genes in tumor and normal tissues.FigureS5.KM estimates of OS in cervical cancer patients with different expression levels of 10 PCD-related genes.FigureS6.Association between programmed cell death-index (PCDi) score and clinical features in cervical cancer patients.FigureS7.KM estimates of OS in patients with different PCDi scores for a particular FIGO cancer stage.FigureS8.KM estimates of OS in patients with different PCDi scores and FIGO cancer stage.FigureS9.Multivariate Cox regression analysis in different study cohorts.FigureS10.The expression profiles of different tumor immune infiltrates between PCDi-High and PCDi-Low groups through ssGSEA algorithm.FigureS11.Violin and scatter plots displaying the association between PCDi and T cell dysfunction potential and exclusion potential.FigureS12.MMP1 expression levels in different cancer types.
Figure S1.Kaplan-Meier (KM) estimates of overall survival (OS) in cervical cancer patients with different FIGO cancer stages.FigureS2.Sample inclusion and exclusion criteria of this study.FigureS3.Univariate Cox regression analysis identified 27 PCD-related candidate genes.FigureS4.Comparison of expression levels of 12 PCD-related genes in tumor and normal tissues.FigureS5.KM estimates of OS in cervical cancer patients with different expression levels of 10 PCD-related genes.FigureS6.Association between programmed cell death-index (PCDi) score and clinical features in cervical cancer patients.FigureS7.KM estimates of OS in patients with different PCDi scores for a particular FIGO cancer stage.FigureS8.KM estimates of OS in patients with different PCDi scores and FIGO cancer stage.FigureS9.Multivariate Cox regression analysis in different study cohorts.FigureS10.The expression profiles of different tumor immune infiltrates between PCDi-High and PCDi-Low groups through ssGSEA algorithm.FigureS11.Violin and scatter plots displaying the association between PCDi and T cell dysfunction potential and exclusion potential.FigureS12.MMP1 expression levels in different cancer types.

Figure S4 .
Figure S4.Comparison of expression levels of 12 PCD-related genes in tumor and normal tissues.

Figure S5 .
Figure S5.KM estimates of OS in cervical cancer patients with different expression levels of 10 PCD-related genes.

Figure S6 .
Figure S6.Association between programmed cell death-index (PCDi) score and clinical features in cervical cancer patients.(A).Heatmap displaying the expression levels of 10 PCD-related genes in relation to clinical characteristics.(B-G).Violin plots depicting the association between normalized PCDi and clinical characteristics, including FIGO cancer stage (B), T stage (C), N stage (D), M stage (E), overall survival status (F), and disease-free survival status (G).

Figure S8 .
Figure S8.KM estimates of OS in patients with different PCDi scores and FIGO cancer stage.

Figure S10 .
Figure S10.The expression profiles of different tumor immune infiltrates between PCDi-High and PCDi-Low groups through ssGSEA algorithm.

Figure S11 .
Figure S11.Violin and scatter plots displaying the association between PCDi and T cell dysfunction potential and exclusion potential.(A).CGCI-HTMCP-CC dataset.(B).GSE52904 dataset.

Figure S12 .
Figure S12.MMP1 expression levels in different cancer types.The expression data of MMP1 was obtained from GEPIA2 database.